SMILES Profiling

Comprehensive SMILES profiling through SwissTargetPrediction, PubChem, ADMETlab 3.0, ChEMBL, and PK-Smart. Use when given a single SMILES to extract predicte...

Security Scan

The skill's code and instructions mostly match its stated purpose (profiling a SMILES string via public tools) but there are inconsistencies (missing dependency/installation guidance and contradictory autoresearch constraints) that could cause unexpected or opaque behavior.

Purpose & Capability

Name/description match the implementation: the scripts and SKILL.md call SwissTargetPrediction, PubChem, ADMETlab 3.0, ChEMBL, and PK‑Smart exactly as advertised. Network calls and parsing logic align with the stated goal of producing a combined pharmacology profile.

Instruction Scope

Runtime instructions focus on the SMILES-to-profiles workflow and reference the included scripts. The scripts perform only the expected network lookups and parsing; they do not attempt to read unrelated system files or request credentials. However, the code will transmit user-provided SMILES to multiple external services (expected for this skill but privacy-relevant) and uses a Streamlit/WebSocket flow to interact with PK-Smart.

Install Mechanism

This is an instruction-only skill with included Python scripts but no install spec or dependency list. The code imports third-party packages (e.g., websockets, google.protobuf) that are not declared anywhere, and autoresearch.config.md states 'Do not add third‑party Python packages' which conflicts with actual imports. That mismatch can lead to runtime failures or hidden ad-hoc installation steps by an integrator; lack of dependency declaration is an operational and supply-chain concern.

Credentials

No environment variables, credentials, or config paths are requested. All external access is to the public APIs/URLs documented in the references. There are no unrelated secrets requested.

Persistence & Privilege

The skill is not force‑enabled (always:false) and does not request to modify other skills or global agent settings. Autonomy (model invocation) defaults are unchanged; nothing in the package requests elevated persistence.

What to consider before installing

This skill appears to do what it says (send a SMILES string to public cheminformatics services and aggregate results), but be aware of two practical risks: (1) the included Python scripts rely on third‑party packages (websockets, google.protobuf, etc.) but the skill provides no install or dependency manifest — installation may fail or require ad-hoc package installs; (2) the skill sends the SMILES (which may be proprietary or sensitive) to multiple external services over the network (PubChem, SwissTargetPrediction, ADMETlab, ChEMBL, PK‑Smart). If you need to protect chemical confidentiality or operate in an air‑gapped environment, do not run this skill without validating where data will be sent and ensuring required libraries are installed in a controlled manner. If you want to proceed, ask the maintainer for a clear requirements.txt or packaging instruction and confirm whether you accept transmitting SMILES to the documented endpoints.

Turn one SMILES into a compact pharmacology profile.

Tool stack

Tool	Use
SwissTargetPrediction	Predicted protein targets with probabilities
PubChem	Exact CID, synonyms, MW, XLogP, TPSA, HBD/HBA, rotatable bonds
ADMETlab 3.0	ADMET flags, drug-likeness, toxicity sections
ChEMBL	Similarity hits, max phase, known mechanisms, target names
PK-Smart	Human and animal PK estimates (VDss, CL, t½, fu, MRT)

Workflow

1. Use the user-provided SMILES as-is unless it clearly needs cleanup.
2. Normalize only obvious salts/counterions.
3. Run SwissTargetPrediction for target hypotheses.
4. Run PubChem for identity and physicochemical baseline.
5. Run ADMETlab 3.0 for ADMET flags and drug-likeness.
6. Run ChEMBL for similarity hits and mechanism context.
7. Run PK-Smart last for PK estimates.
8. Merge the outputs into one short report, preferring direct database evidence over predictions.

Preferred scripts

• scripts/run_smiles_profile.py <SMILES> for the full report.
• scripts/run_smiles_smoke.py --cases <tsv> for regression/benchmark runs.

Extraction notes

• SwissTargetPrediction: capture the top 5 targets, probabilities, and species.
• PubChem: capture CID, IUPAC name, synonyms, MW, XLogP, TPSA, HBD/HBA, and rotatable bonds.
• ADMETlab 3.0: capture the section list and any obvious toxicity/drug-likeness flags.
• ChEMBL: capture the similarity threshold, top hits, max phase, mechanism_of_action, and target name.
• PK-Smart: capture VDss, CL, t½, fu, MRT, and animal PK when available.

Handling edge cases

• Salts/mixtures: strip common counterions first.
• Invalid SMILES: fail cleanly, do not crash.
• Missing tools: continue with the remaining sources.
• PubChem miss: report the miss and continue.
• ChEMBL miss: automatically fall back from 90 → 80 → 70 similarity.

Output shape

SMILES: CC(=O)OC1=CC=CC=C1C(=O)O

## SwissTargetPrediction
- Top target: PTGS1 (prob: 0.76)

## PubChem
- CID: 2244 | Name: 2-acetyloxybenzoic acid
- MW: 180.16 Da | XLogP: 1.2 | TPSA: 63.6 Ų

## ADMETlab 3.0
- MW: 180.04 Da | QED: N/A

## ChEMBL
- Top hit: CHEMBL25 ASPIRIN | sim 100 | phase 4
- Mechanism: Cyclooxygenase inhibitor -> Cyclooxygenase

## PK-Smart
- VDss: 0.23 L/kg | CL: 10.41 mL/min/kg
- t½: 0.34 hr | fu: 0.67